real exchanges rates in commodity producing countries: a ...signi cant positive long-run impact on...

Real Exchanges Rates in Commodity Producing Countries: A Reappraisal

V. Bodart, B. Candelon and J-F. Carpantier

Discussion Paper 2011-7

Real Exchanges Rates in Commodity Producing Countries: A

Reappraisal

Vincent Bodart1 Bertrand Candelon2 Jean-Francois Carpantier3

Version: 17 February 2011

Abstract

Commodity price booms, as those recorded in the last decade, may have a signifi-

cant economic impact in small, commodity exporting, developing countries. Whether the

impact on output is positive or negative is still unclear. It depends on various factors,

notably on the impact that commodity prices can have on the real exchange rate of the

commodity exporting countries. Two recent papers show that the real exchange rate ap-

preciates when commodity prices increase. Our analysis produces new estimates of this

relationship by focusing on a large sample of developing countries which are specialized

in the export of one leading commodity. By using non-stationary panel techniques robust

to cross-sectional-dependence, we find that the price of the dominant commodity has a

significant long-run impact on the real exchange rate when the exports of the leading com-

modity have a share of at least 20 percent in the country’s total exports of merchandises.

Our results also show that the larger this share, the larger the size of the impact.

Keywords:

Commodity producers, Commodity prices, Natural resource curse, Non-stationary panel,

Real exchange rates

1Universite Catholique de Louvain, Department of Economics and IRES, B-1348 Louvain-La-Neuve, Bel-

gium.2University of Maastricht, Department of Economics, PO Box 616, 6200 MD Maastricht, The Netherlands.

Corresponding author. [email protected] Catholique de Louvain, Department of Economics, CORE and IRES, B-1348 Louvain-La-Neuve,

Belgium.

1

1 Introduction

Many developing countries, gifted in natural resources, heavily depend on international com-

modity prices. Tobacco, for example, represents about 60% of Malawi commodity exports, oil

90% of Nigeria exports. Whether natural resources are a luck or a curse for these countries is

still an open debate. Some countries have admirably well taken profits of their endowments,

while others suffered countless economic difficulties. Whatever the records, it is generally

recognized that commodity prices can be a source of macroeconomic instability in developing

countries.

The natural resource curse literature, surveyed recently by Frankel (2010a), identifies several

channels through which commodity prices can affect the economic situation of commodity

producing countries. One important channel is the real exchange rate. Theoretical literature

indeed shows that commodity price booms may cause an appreciation of the real exchange

rate of commodity exporting countries. This appreciation in turn generates so-called “Dutch

disease” effects, by altering the competitiveness of the non-commodity exportable sectors.

In this paper, we explore this “real exchange rate” channel and we provide new empirical

estimate of the relationship between commodity prices and real exchange rates for a large

group of developing countries. Recent estimates of this relationship are provided by Chen

and Rogoff (2003) and Cashin et al. (2004). The former provides results for 3 developed

countries (Canada, Australia and New-Zealand) and the latter for several developing coun-

tries. Both use time-series cointegration techniques and find that commodity prices have a

significant positive long-run impact on the real exchange rate of many countries.

In their analysis, Chen and Rogoff (2003) and Cashin et al. (2004) look at the impact that

commodity prices can have on real exchange rates by using for each country a commodity

price index constructed as a weighted average of the price of the different commodities that are

produced and exported by the country. Accordingly, they do not pay a particular attention to

countries that are specialized in the export of one leading commodity, as for instance Burundi

whose coffee exports count for about 50% of its total exports of merchandises or Mali where

gold counts for more than 54% of its total exports. For several reasons, we do believe how-

ever that it is relevant to investigate the particular case of countries highly specialized in the

2

export of one leading commodity and determine what is the impact of the price of the leading

commodity that they export on their real exchange rate. First, a high specialization implies

that commodity price fluctuations lead to large variations in the external trade balance. That

is the reason why Frankel and Saiki (2002) made the proposal that countries specialized in the

export of a particular commodity should peg their currency to the price of that commodity.

This mechanism would permit an automatic accommodation of terms of trade shocks. Second,

the use of an aggregate price export index instead of a single price implies that variations in

the price index reflect not only changes in the individual commodity prices but also changes in

the respective weights of the commodities. In addition, when an aggregate price index is used,

the relationship that is estimated between the real exchange rate and the commodity price in-

dex depends on the correlation among individual prices over the sample period. Accordingly,

the true relationship between real echange rates an commodity prices is blurred. For instance,

if the aggregate price index is composed of two prices that are perfectly negatively correlated,

it is impossible to detect any relationship between the real exchange rate and the aggregate

price index. One way to avoid these problems is to use a single commodity price instead of

an aggregate price index. For all these reasons, contrary to previous research, our analysis

will focus on developing countries that are specialized in the export of one leading commodity.

Several previous papers have been interested in the determination of the real exchange rate

of countries with a strong export specialization. This is for instance the case of Edwards

(1986), who studies the determination of the real exchange rate in Colombia, a world leading

exporter of coffee; of Habib and Kalamova (2007), who concentrate on oil exporting countries;

of Sjaastad and Scacciavillani (1996), whose analysis focuses on gold exporting countries; or

of Frankel (2007), who investigate whether the price of gold affects the real exchange rate of

the South-African rand. Our research is the first, to our knowledge, to examine this issue for

a large sample of developing countries.

Our analysis of the relationship between real exchange rates and commodity prices relies

non-stationary panel cointegration methods of second generation, i.e. which are robust to

cross-sectional dependence. Anticipating on our results, we find that the price of the leading

commodity is a long-run determinant of the real exchange rate of countries where the main

3

commodity accounts for at least 20 percent of the total merchandise exports of the country.

We also find that the higher the specialization of the country, the larger the elasticity.

We also show that the estimate of the cointegration relationship between real exchange rates

and commodity prices is markedly different whether the econometric technique that is used

is robust or not to cross-section dependence. Real exchange rates are by construction inter-

dependent and commodity prices share some co-movements, as confirmed by Pesaran (2004)

tests. By comparing estimates obtained from robust and non-robust techniques, from Bai et

al. (2009) on the one hand and fully modified OLS and dynamic OLS on the second hand, we

find that the elasticity of real exchange rates to commodity prices is strongly overestimated

when methods not robust to cross-sectional dependence are used, as it is the case in Coudert,

Couharde and Mignon (2008).

The rest of the paper is organized as follows. Section 2 presents briefly the empirical literature

devoted to the relation between commodity prices and real exchange rates. Section 3 presents

the model. Section 4 describes the data. Section 5 presents the methods. Section 6 discusses

the empirical results. Some robustness checks are performed in section 7. Section 8 concludes.

2 Review of literature

The determination of real exchange rates has always been a topic of strong interest among

academics and policymakers. For economies that are largely open, as small developing coun-

tries are, the real exchange rate is indeed a key economic variable. For instance, it can be used

to assess easily the competitive position of a country; it helps to detect undesirable distortions

in the factor allocation such as in the Dutch disease; it can also play an important role in the

emergence of large current account imbalances. For a recent discussion on the importance of

real exchange rates, see Chinn (2006).

Purchasing power parity (PPP) is the basic model of exchange rate determination. It states,

in its weak version, that differentials of inflation are neutralized by a corresponding adjust-

ment of the nominal exchange rate. Therefore, if PPP holds, shocks to real exchange rates

should be transitory and real exchange rates should revert over time to their long-run mean.

4

If, on the contrary, shocks are permanent, PPP is rejected. Though its theoretical appeal

and a voluminous empirical literature, the empirical support to PPP is mixed (see Rogoff

(1996) for a survey) and the idea that equilibrium exchange rates are non stationary is now

largely admitted. Variations in equilibrium real exchange rates are mostly attributed to cu-

mulated current account imbalances, government spending, real interest rate differentials,

sectoral productivity shocks (the ”Balassa-Samuelson” effect), natural resources discovery,

and terms-of-trade shocks.

For commodity producing countries, terms-of-trade shocks induced by changes in world com-

modity prices constitute a potential key determinant of their equilibrium real exchange rates.

This is particularly true for small developing countries, where primary commodities represent

a large share of their exports. Recent empirical evidence provide support to this view. Chen

and Rogoff (2003) show for instance that commodity prices have a strong and positive influ-

ence on the real effective exchange rate of Canada, Australia and New-Zealand. Similarly,

Cashin et al. (2004) search for a long-run equilibrium relationship between real commodity

prices and the real effective exchange rate of 58 commodity-exporting countries and found

evidence of such a relationship for about one-third of their sample of countries. Coudert et al.

(2008) report similar evidence for a large group of commodity exporting countries, including

oil producers.

Following the research initiated by Chen and Rogoff (2003) and Cashin et al. (2004), several

recent papers have provided new evidence about the long-run impact of commodity prices on

real exchange rates. For instance, Koranchelian (2005) examines the relationship between oil

prices and the real exchange rate of Algeria. Habib and Kalamova (2007) investigate the same

relationship but for several oil producing countries. Coudert et al. (2008) consider a large set

of countries, one of their objectives being to see whether the long-run impact of commodity

prices on real exchange rates differs between oil exporters and non-oil commodity exporters.1

In the most recent papers, for instance Coudert et al. (2008) or Chen and Chen (2007), panel

cointegration techniques are used, whereas the results of Chen and Rogoff (2003) and Cashin

et al. (2004) were obtained with time-series techniques.

5

Our analysis of the long-run relationship between real exchange rates and commodity prices

differs from previous ones in two respects. First, for the reasons explained in the introduction,

the real exchange rate of every country of our sample is related to the price of the commodity

that dominates the exports of the country. Second, cointegration relationships are tested and

estimated using non-stationay panel methods robust to cross-sectional dependence.

3 Model

Our model consists of a simple univariate relationship between real exchange rates and com-

modity prices. Formally, we have:

REERi,t = αi + βCOMi,t + εi,t (1)

where REERi,t is the real effective exchange rate (in logarithm) of country i, COMi,t the

price of leading export commodity (in logarithm) of country i and the error term εi,t is i.i.d.

over periods but correlated across cross-sectional units. As in Chen and Rogoff (2003) or

Cashin, Cespedes and Sahay (2004), our model only includes a single regressor. This is mo-

tivated by the fact that several traditional explanators of real exchange rates are considered

as being irrelevant for small developing countries. For instance, given that many small de-

veloping countries are poorly integrated to world financial markets, it is very unlikely that

real interest rate differentials or net foreign asset accumulation be a significant determinant

of real exchange rates. Furthermore, should some of these other determinants be relevant,

like the Balassa-Samuelson effect, the data are very often not available or of low quality. In

any case, the non-stationary panel approach that we use guarantees that our results be at

least consistent.

Relationship (1) will be estimated using panel cointegration methods with monthly data cov-

ering the period 1988-2008.

6

4 Data

We use three datasets. Commodity exports come from the UN Comtrade database. These

data are necessary to make the selection of countries that are specialized in the export of

one particular commodity. Commodity prices are taken from the IMF International Finan-

cial Statistics database. The CPI-based real effective exchange rates come from the IMF

Information Notice System database.

4.1 Selection of the Relevant Commodity-Country Pairs

In order to detect the countries and commodities eligible for our analysis, we recovered from

the UN Comtrade database the annual US$ export value of 42 commodities for 68 countries

over 21 years (1988-2008). Our sample of commodities is similar to the one of Cashin et al.

(2004), with the addition of oil. The list of commodities is reported in Table 1. The countries

included in our sample are the developing and emerging countries selected by Cashin et al.

(2004), on the basis of the IMF classification of commodity countries. No advanced countries

are included in our sample. The list of countries is reported in Table 2.

As explained in the previous sections, what is new in our approach is that we relate the real

exchange rate of an individual country to the price of the commodity which is the dominant

export of that country. To do so, we have to identify from our set of countries and commodities

a series of country-commodity pairs. We proceed to the construction of that series as follows:

1. For every country in the sample, we compute the 1988-2008 average ratio of the export

value of each commodity exported by the country in the total value of all commodity

exports. For example, in the case of Mali, the average share of its exports of cotton in

its total exports of all commodities is 33%.

2. Using these commodity export ratios, we then determine for each commodity which

country is the most highly specialized in the export of that particular commodity. We

then use the country that is so identified to obtain one particular country-commodity

pair. For example, this procedure shows that Mali has the highest ratio of cotton exports

among the countries in our sample. We applied this procedure to the 42 commodities

of our sample and we thus obtained a series of 42 country-commodity pairs.

7

3. We then eliminate from the series the pairs where the commodity share is less than 20%

of total commodity exports. This threshold is set arbitrarily but other thresholds will

be used as robustness checks.

4. It appeared that, for some countries, the export share of the dominant commodity was

very volatile over time. We then decided to eliminate a country-commodity pair when

the share of its dominant export was less than 2% during 5 years or more. We removed

for instance Mozambique, because the share of its aluminium exports in total exports

was more than 50% after 2001 but about zero before 2000.

In Table 3, we report for several commodities the 3 countries with the largest commodity

share. Commodities not having a share of 5% in at least one country are not reported in the

Table. Following the selection procedure just described, 11 country-commodity pairs are kept

for our empirical analysis. They are listed in Table 4. The pair Cotton-Benin is not reported

due to the unavailability of real exchange rate data over the period 1980-2008. In the last

section, we do perform robustness checks using alternative samples.

An important feature about the construction of our sample of countries is that only one

country is associated to each commodity. It follows that some countries which are highly

specialized in the export of a particular commodity, as for instance the OPEC countries, are

not included in our sample. By proceeding like this, we avoid to give too much weight to some

commodities, like oil, and so make sure that our results are not influenced by a few number

of commodities.

4.2 Commodity Prices

Commodity prices are taken from the International Financial Statistics (IFS) database of the

IMF. Two series, Tobacco and Gold, were not available in IFS and were taken from Datas-

tream (with respective codes :USI76M.ZA and GOLDBLN). We provide details of each series

in Table 5. All the series are available at the monthly frequency over the period 1980-2008.

In order to capture properly the relationship between real exchange rates and commodity

prices, commodity prices have to be expressed in real terms. Following Cashin, Cespedes and

8

Sahay (2004), we compute the real price of each commodity by deflating the price of each

commodity by the IMF’s index (of the unit value) of manufactured exports (MUV).2 The

use of the MUV index as a deflator is common in the commodity-price literature (see for

example Deaton and Miller (1996) and Cashin et al. (2004)). The MUV index is represented

in Figure 3. Real commodity prices are then normalized with base period January 1995=100.

Normalized real commodity prices are represented in Figure 1.

We can see in Figure 1 that real commodity prices (especially gold, oil, uranium and copper)

wandered around a long run average, or slightly decreased over time with a sudden and

steep boost during last years. Banana on the contrary exhibits large fluctuations, with some

potential seasonality pattern.

4.3 Real Exchange Rates

As in common in many studies on the determination of real exchange rate, real exchange

rates are real effective exchange rates based on consumer prices.3 The data come from the

IMF’s Information Notice System (INS).4

We represent the CPI-based real effective exchange rates (normalized to January 1995=100)

in Figure 2 for the period 1980 to 2008. We can see that some countries experienced sudden

depreciation. For example, Zambia experienced some difficult spells during the middle of the

eighties when the Zambian currency fell abruptly in the context of unsuccessful IMF inter-

ventions. In addition, three countries in our selection are part of the CFA Franc zone, whose

currency was devaluated by 50% against the French Franc in January 1994. Note also that

the CFA Franc is linked to the euro since 1999.

To anticipate the econometric analysis, the real exchange rate and the real commodity price

for our sample of 11 country-commodity pairs are plotted in Figures 4 and 5. One can observe

that some series seem to be clearly correlated, while others are not. Formal statistical tests

are performed later.

9

4.4 Descriptive Statistics

Some elementary statistics relative to the real exchange rates and the real commodity prices

(both expressed in logs) are reported in Table 6. We see that there are big differences in the

standard deviation across commodity prices. For example, the standard deviation of tobacco

prices is more than nearly ten times lower than the one of uranium. We may also observe

differences in the variability of real exchange rate across countries, the standard deviation

of the real exchange rate of Nigeria being more than ten times larger than the one of Do-

minica. We find unit roots for most of the series (excepted banana, soya and tea prices) using

Dickey-Fuller tests (with constant and trend, or just a constant if the trend is not significant).

5 Econometric methodology

In the literature, due to the potential non-stationarity of the series, the most common method

for analyzing the dependence of the real exchange rate of highly specialized countries on the

international price of their dominant commodity is related to cointegration methods (see for

example Cashin et al. (2004) who investigated a similar problem but with country-specific

commodity indices). Although time-series cointegration methods could be used to test the

relationship between real exchange rates and real commodity prices, as in Chen and Rogoff

(2003) and Cashin et al. (2004), we considered that these methods were not relevant for our

application because the number of observations per country was too small to guarantee that

the unit-root based tests reach a good power. We therefore use panel methods to perform

our econometric analysis.

Techniques combining panel and non-stationary series give rise to three kinds of methods

and tests: panel unit root tests, panel cointegration tests and cointegration estimation and

inference. One can distinguish two generations of panel methods. In the first generation, the

methods are based on the assumption that panel units are cross-sectionally independent. Re-

garding our dataset, this assumption amounts to consider, for instance, that the real exchange

rates of Mali and Kenya are independent, as would be cocoa and coffee prices. It is obvious

that such an assumption is unrealistic for effective exchange rates and commodity prices, as it

10

is confirmed formally by the results of the cross-sectional dependence test of Pesaran (2004)

reported in Table 7. We therefore use second generations panel techniques, which allow for

cross-sectional dependence. We briefly describe these methods in the following subsections.5

5.1 Panel Unit Roots Tests

In the empirical analysis, we first test for the non-stationarity of real exchange rates and of

commodity prices. First generation panel unit root tests proposed by Levin, Lin and Chu

(2002) (LLC hereafter) and Im, Pesaran and Shin (2003) (IPS hereafter) are the most popular

tests in empirical studies. LLC pool the panel series, correct and standardize the t-stat to

make it normally distributed. IPS on the other hand do not pool the data but estimate sep-

arate augmented Dickey-Fuller unit root tests for cross-section units and average the t-tests.

After standardization, this average follows a normal distribution. Though these tests are

widely applied, it has been shown that they are inconsistent in the presence of cross-sectional

dependence, as well as when N (the cross-sectional dimension) is small with respect to T (the

time dimension). Several alternatives and modifications of these tests have been proposed

recently (see Hurlin and Mignon (2007), Breitung and Pesaran (2005) and Gengenbach et al.

(2010) for recent reviews of panel unit toot tests).

The second generation technique that we use is based on the subsampling approach, proposed

by Choi and Chue (2007). This approach is applied to LLC and IPS and provides results

robust to cross-section dependence. The idea of this approach is to approximate the limiting

distribution of the tests by computing tests with smaller blocks of consecutively observed

time series and to compute the empirical distribution. Hence, this method does not require

estimation of nuisance parameters. As the determination of the block size may be based

on two rules (stochastic calibration or minimum volatility), we will present results for both

methods. In addition to the LLC and IPS tests, we also present results for an alternative

test, labeled as “inverse normal panel unit root test” (INVN) (Choi 2001), which corresponds

to a generalized least squares version of the ADF test.

11

5.2 Panel Cointegration Tests

If panel unit root tests do not reject the hypothesis that real exchange rates and commodity

prices are non stationary, the next step is to check whether the series are cointegrated. If

it is the case, this means that real exchange rates of highly specialized countries are tied to

commodity prices of their dominant commodity. In a panel context, the usual tests are those

developed by Kao (1999) and Pedroni (1999). These tests belong to the residuals based cointe-

gration tests family. These techniques rely on the assumption of cross-sectional independence.

In order to obtain results that are robust to the cross-sectional dependence, we follow the

method proposed by Fachin (2007), which consists in using a block-bootstrap version of the

group and the mean t-statistic of Pedroni (1999). His method relies on fast-double bootstrap

procedures of Davidson and MacKinnon (2000) which combine good size and power prop-

erties with reasonable computing power requirements. The theoretical validity of bootstrap

procedures in the non-stationary panel context have recently been developed by Palm et al.

(2008).

5.3 Panel Cointegration Estimate and Inference

If real exchange rates and commodity prices are found to be non-stationary, the next step is

to estimate the cointegration coefficient β in equation 1.

As it is well known, the use of normal OLS techniques leads to spurious regression when the

series are non stationary and, consequently, specific panel-cointegration techniques have to be

used. Phillips and Moon (2000) show that in the case of homogeneous and near-homogeneous

panels, the coefficient of cointegration can be estimated by a fully modified (FM) estimator.

This method is non-parametric as it employs kernel estimators of the nuisance parameters

that affect the asymptotic distribution of the OLS estimator. It tackles the possible problem

of endogeneity of the regressors as well as the autocorrelation of residuals. Alternatively,

Kao and Chiang (2000) and Mark and Sul (2003) propose a dynamic least square estimator

(DOLS). This estimation procedure is parametric and has the advantage of computing con-

venience.

12

Though largely used in the empirical literature, these techniques have a major weakness

since they assume cross-section independence. Because of this, we use the technique recently

developed by Bai, Kao and Ng (2009) (BKN), whic is robust to cross-sectional dependence.

They consider the following framework:

REERi,t = βCOMi,t + ei,t (2)

COMi,t = COMi,t−1 + εi,t (3)

ei,t = λ′iFt + ui,t (4)

Ft = Ft−1 + ηi,t, (5)

where Ft are unobserved factors and λi the factor loadings. The BKN method imposes a factor

structure on ei,t to capture the cross-sectional dependence. They use an iterative procedure

to estimate jointly the factors and the cointegration coefficient (β). We present the results

obtained with the three estimators (BKN, FMOLS and DOLS) and compare them.

6 Results

In this section, we successively test the non-stationarity of real exchange rates and of com-

modity prices, and the existence of a single homogeneous cointegration relationship between

them. We then estimate the cointegration relationship.

6.1 Stationarity analysis

Output of the stationarity tests of commodity prices is reported in Table 8. We observe that

each of the three tests considered (LLC, IPS and INVN) do not reject the null hypothesis

of unit root in the panel. This result holds at the 10% level of significance. It is verified

whether stochastic calibration or minimum volatility block selection rules are used (see Choi

and Chue (2007) for details). In only one case, namely the IPS test with stochastic calibra-

tion rule, the null of non-stationarity is rejected. Therefore, there is enough evidence to treat

commodity prices as non-stationary. This result is interesting as it departs from the usual

idea that commodity prices are partially predictable due to seasonality patterns (Gorton and

13

Rouwenhorst, 2004).

Panel unit root tests for the real exchange rates are reported in Table 9. Without any

exception, all the tests (LLC, IPS, INVN, with either SC or MV block selection rule) confirm

that real exchange rates possess a panel unit root. PPP is therefore rejected.

6.2 Cointegration analysis

Results of Fachin (2006)’s panel cointegration tests are reported in Table 10. We report both

mean and median t-tests, based on Pedroni (1999), with and without time dummies. All

the tests consistently reject the null hypothesis of no cointegration at the 5 percent level.

Thus, the presence of a long-run relationship between the real exchange of countries highly

specialized in the export of one particular commodity and the price of that leading commodity

is suppoted by our data. Treating each country individually, Cashin et al. (2004) also find

numerous cases where real exchange rates and commodity prices are cointegrated but, in

their study, a country specific commodity price index is used rather than the price of a single

commodity as it is the case here.

6.3 Estimation of the cointegration relationship

Following the results of the previous subsection, we now proceed to the estimation of the

cointegration relationship. We compare the results obtained with first generation techniques

(DOLS and FMOLS) to the estimates provided by BKN which, as discussed before, is robust

to cross-section dependence. Results are reported in Table 11.

Whatever technique is used (DOLS, FMOLS, BKN), it tunrs out that the cointegration coef-

ficient is significant and positive, as expected. The coefficient estimates obtained with DOLS

and FMOLS are respectively 0.326 and 0.317, which suggests that commodity price shocks

have a strong long-run impact on the real exchange rate of highly specialized countries. For

example, according to these results, the real exchange rate of an oil producing country should

increase by 3% when the international price of oil increases permanently by 10%. Our coeffi-

cient estimate is however smaller than the 0.6484 coefficient found by Coudert et al. (2008).

They also use DOLS but their regressor is a country-specific commodity price index rather

14

than the price of the dominant commodity.

When the BKN methodology is used, the estimate of the cointegration coefficient is 0.149, thus

much smaller than the coefficient obtained with DOLS and FMOLS. By correcting the bias

induced by cross-section dependence, we thus find that there is still a positive and significant

long-run impact of commodity prices on the real exchange rate of highly specialized countries,

but the impact is now economically small. Our results also show that the results found in

earlier studies with DOLS and FMOLS estimates are overestimated.

7 Robustness checks

To verify whether our results are specific to the selection of countries that we made, we repli-

cated the cointegration analysis and the BKN estimation for other selections of countries. So

far, a country has been considered as having a dominant commodity export if the share of

that particular commodity in its total exports was at least 20%. Following the methodology

presented in section 4.1, we constructed five new samples of countries by fixing the minimum

export share at 50%, 40%, 30%, 15% and 10%. The samples get larger as the threshold

decreases: 5 countries for 50%, 6 for 40%, 9 for 30%, 12 for 15% and 14 for 10%. Results

obtained with these new samples are reported in Table 13. They show that cointegration

holds from a threshold of about 20%. Indeed, when the threshold is set at 10% and 15%,

we cannot reject the null of no cointegration between the commodity price of the leading

commodity and the real exchange rate (see Table 12.) In addition, the results in Table 13,

which are illustrated in Figure 6, show that the larger the share of the leading commodity in

total exports, the larger the elasticity between real exchange rates and commodity prices.

We found in Table 11 that DOLS and FMOLS methods tend to overestimate the cointegration

coefficient, compared to the BKN method, which is robust to cross-section dependence. We

reiterated our estimates with the samples obtained by setting the threshold at 50%, 40% and

30%, which are the cases where cointegration cannot be rejected. The results are reported in

Table 14 and confirm that neglecting cross-sectional dependence leads to a higher elasticity

estimation.

15

We also assessed the robustness of our results to the presence of structural breaks in the

series of real exchange rates (resulting for instance from large devaluations). We consider as

an outlier a real exchange rate monthly variations larger than 50% (see Table 15). Outliers

are removed by simply setting the monthly change equal to zero. We then reiterate our tests

with the modified dataset. As reported in Tables 16 and 17, the results are similar to those

obtained with the original dataset. To further check the potential impact of the breaks, we

reiterate our estimates by eliminating as above the monthly variations of the real exchange

rates larger than 50%, but also the January 1994 monthly variations for Mali, Cote d’Ivoire

and Niger due to the devaluation of the CFA Franc. Our results remain similar as reported

in Tables 16 and 17.

8 Conclusion

The last decade has seen a sustained increase in the price of agricultural and mineral com-

modities. Whether commodity price booms have a positive or negative impact on the output

of commodity exporting countries remains unclear so far. One particular reason why the im-

pact could be negative is an appreciation of the real exchange rate in response to the increase

of commodity prices. In this paper, we explored this relationship between real exchange rates

and commodity prices for a set of developing countries specialized in the export of one leading

commodity. We show that the real exchange rate appreciates when the price of the leading

commodity exported by the country increases, provided that the dominant commodity ac-

counts for at least 20 percent of the total exports of the country. We also showed that the

larger the share of the main exported commodity, the stronger is the impact on the real ex-

change rate.

Our results therefore suggest that small developing countries heavily specialized in the export

of one commodity are vulnerable to “Dutch disease” effects. One way to prevent the emer-

gence of these effects would be to ease monetary (or exchange rate) policy when there is a

long and lasting increase in commodity prices.

Notes

1. See also Chen and Chen (2007) for the relation between oil and importing-countries real

16

exchange rates.

2. MUV is the unit value index (in US dollars) of manufacturing exports from 20 developed

countries with country weights based on the countries’ total 1995 exports of manufactures

(base 1995=100). This MUV index deflator is provided by the IMF’s IFS database.

3. An alternative real exchange rate, based on unit labour cost, is available only for a few

countries of our sample.

4. See Desruelle and Zanello (1997) for details regarding the construction of the real effective

exchange rates.

5. See the following surveys for a more detailed review of these techniques: Baltagi and Kao

(2000), Breitung and Pesaran (2005), Hurlin and Mignon (2006, 2007) and Bai et al. (2009)

for the inference technique.

References

[1] Bai J., Kao C., Ng S., Panel Cointegration with Global Stochastic Trends, Journal of

Econometrics, 149, 82-99, 2009

[2] Baltagi B. H., Kao C., Nonstationary Panels, Cointegration in Panels and Dynamic Panels:

A Survey, Center for Policy Research, Maxwell School, Syracuse University, WP 2000-16,

2000

[3] Breitung J., Pesaran M., Unit Roots and Cointegration in Panels, Institute of Economic

Policy Research, WP 05-32, 2005

[4] Cashin P., Cespedes L.F., Sahay R., Commodity Currencies and the Real Exchange Rate,

Journal of Development Economics, 75, 239-268, 2004

[5] Chen S., Chen H., Oil Prices and Real Exchange Rates Energy Economics, 29, 390-404,

2007

[6] Chen Y., Rogoff K., Commodity Currencies, Journal of International Economics, 60, 133-

160, 2003

17

[7] Chinn M., A Primer on Real Effective Exchange Rates: Determinants, Overvaluation,

Trade Flows and Competitive Devaluation Open Economies Review, 17, 115-143, 2006

[8] Choi I., Unit Root Tests for Panel Data, Journal of International Money and Finance, 20,

249-272, 2001

[9] Choi I., Chue T. K., Subsampling Hypothesis Tests for Nonstationary Panels with Ap-

plications to Exchange Rates and Stock Prices, Journal of Applied Econometrics, 22(2),

233-264, 2007

[10] Coudert V., Couharde C., Mignon V., Do Terms of Trade Drive Real Exchange Rates?

Comparing Oil and Commodity Currencies, CEPII research center, WP 2008-32, 2008

[11] Davidson R., MacKinnon J. G., Improving the Reliability of Bootstrap Tests with the

Fast Double Bootstrap, Queen’s University, Department of Economics, WP 995, 2000

[12] Deaton A., Miller R., International Commodity Prices, Macroeconomic Performance and

Politics in Sub-Saharan Africa, Journal of African Economies, 5, 99-191, 1996

[13] Desruelle D., Zanello A., A Primer on the IMF’s Information Notice System, Interna-

tional Monetary Fund, WP 97/71, 1997

[14] Edwards S., Commodity Export Prices and the Real Exchange Rate in Developing Coun-

tries: Coffee in Colombia, in Economic Adjustment and Exchange Rates in Developing

Countries, National Bureau of Economic Research, 233-266, 1986

[15] Fachin S., Long-Run Trends in Internal Migrations in Italy: a Study in Panel Cointegra-

tion with Dependent Units, Journal of Applied Econometrics, 22, 401-428, 2007

[16] Frankel J. A., On the Rand: Determinants of the South African Exchange Rate, National

Bureau of Economic Research, WP 13050, 2007

[17] Frankel J. A., The Natural Resource Curse: A Survey, National Bureau of Economic

Research, WP 15836, 2010a

[18] Frankel J. A., A Comparison of Monetary Anchor Options, Including Product Price

Targeting, for Commodity-Exporters in Latin America, National Bureau of Economic Re-

search, WP 16362, 2010b

18

[19] Gengenbach C., Palm F., Urbain J., Panel Unit Root Tests in the Presence of Cross-

Sectional Dependencies: Comparison and Implications for Modelling, Econometric Re-

views, 29, 111-145, 2010

[20] Gorton G., Rouwenhorst K. G., Facts and Fantasies about Commodity Futures, National

Bureau of Economic Research, WP 10595, 2004

[21] Habib M.M., Kalamova M. M., Are there Oil Currencies? The Real Exchange Rate of

Oil Exporting Countries, Working Paper Series 839, European Central Bank, 2007

[22] Hurlin C., Mignon V., Une Synthese des Tests de Racine Unitaire sur Donnees de Panel,

WP HAL, 2006

[23] Hurlin C., Mignon V., Second Generation Panel Unit Root Tests, WP HAL, 2007

[24] Im K. S., Pesaran M. H., Shin Y., Testing for Unit Roots in Heterogeneous Panels,

Journal of Econometrics, 115, 53-74, 2003

[25] Kao C., Spurious Regression and Residual-Based Tests for Cointegration in Panel Data,

Journal of Econometrics, 90, 1-44, 1999

[26] Kao C., Chiang M., On the Estimation and Inference of a Cointegrated Regression,

in Panel Data Nonstationary Panels, Panel Cointegration and Dynamic Panels, Elsevier

Science Inc., 2000

[27] Koranchelian T., The Equilibrium Real Exchange Rate in a Commodity Exporting Coun-

try: Algeria’s Experience International Monetary Fund, WP 05/135, 2005

[28] Levin A., Lin C., Chu J.C., Unit Root Tests in Panel Data: Asymptotic and Finite-

Sample Properties, Journal of Econometrics, 108, 1-24, 2002

[29] Mark N. C., Sul D., Cointegration Vector Estimation by Panel DOLS and Long-run

Money Demand, Oxford Bulletin of Economics and Statistics, 65, 655-680, 2003

[30] Palm F. C., Smeekes S., Urbain J., Cross-Sectional Dependence Robust Block Boot-

strap Panel Unit Root Tests, Maastricht Research School of Economics of Technology and

Organization, Research memoranda 048, 2008

19

[31] Pedroni P., Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple

Regressors, Oxford Bulletin of Economics and Statistics, 61, 653-70, 1999

[32] Pesaran M.H., General Diagnostic Tests for Cross Section Dependence in Panels, Cam-

bridge Working Papers in Economics, n435, 2004

[33] Phillips P., Moon H., Nonstationary Panel Data Analysis: an Overview of some Recent

Developments, Econometric Reviews, 19, 263-286, 2000

[34] Rogoff K., The Purchasing Power Parity Puzzle, Journal of Economic Literature, 34,

647-668, 1996

[35] Sjaastad L. A., Scacciavillani F., The Price of Gold and the Echange Rate, Journal of

International Money and Finance, 15(6), 879-897, 1996

20

Table 1: List of commodities (with their HS1992 codes)

Code CommodityH0-0201 Meat of bovine animals, fresh or chilledH0-03 Fish, crustaceans, molluscs, aquatic invertebrates neH0-0306 CrustaceansH0-0803 Bananas, including plantains, fresh or driedH0-0901 Coffee, coffee husks and skins and coffee substitutesH0-0902 TeaH0-1001 Wheat and meslinH0-1005 Maize (corn)H0-1006 RiceH0-120710 Palm nuts and kernelsH0-120810 Soya bean flour or mealH0-1513 Coconut, palm kernel, babassu oil, fractions, refinedH0-17 Sugars and sugar confectioneryH0-18 Cocoa and cocoa preparationsH0-1801 Cocoa beans, whole or broken, raw or roastedH0-24 Tobacco and manufactured tobacco substitutesH0-2401 Tobacco unmanufactured, tobacco refuseH0-2510 Natural phosphates (calcium, calcium-aluminium), chalH0-2612 Uranium or thorium ores and concentratesH0-2701 Coal, briquettes, ovoids etc, made from coalH0-2704 Retort carbon, coke or semi-coke of coal, lignite,peaH0-2705 Coal gas, water gas, etc. (not gaseous hydrocarbons)H0-2709 Petroleum oils, oils from bituminous minerals, crudeH0-2835 Phosphatic compoundsH0-2919 Phosphoric esters, their salts and derivativesH0-40 Rubber and articles thereofH0-4001 Natural rubber and gums, in primary form, plates, etcH0-52 CottonH0-5201 Cotton, not carded or combedH0-7108 Gold, unwrought, semi-manufactured, powder formH0-72 Iron and steelH0-7201 Pig iron and spiegeleisen in primary formsH0-74 Copper and articles thereofH0-7401 Copper mattes, cement copper (precipitated copper)H0-75 Nickel and articles thereofH0-7502 Unwrought nickelH0-76 Aluminium and articles thereofH0-7601 Unwrought aluminiumH0-79 Zinc and articles thereofH0-7901 Unwrought zincH0-80 Tin and articles thereofH0-8001 Unwrought tinNotes.The codes are based on the HS1992 classification used in the UN Com-trade database.

21

Table 2: List of countries

CountriesAlgeria India Papua New GuineaArgentina Indonesia ParaguayBahrain Iran PeruBangladesh Jordan QatarBenin Kenya Saudi ArabiaBolivia Kuwait South AfricaBrazil Madagascar Sri LankaBurundi Malawi SudanCote d’Ivoire Malaysia SurinameCameroon Mali SyriaCentral African Rep. Mauritania TanzaniaChile Mexico ThailandColombia Morocco TogoCosta Rica Mozambique TunisiaDominica Myanmar TurkeyDominican Rep. Nicaragua UgandaEcuador Niger United Arab EmiratesEgypt Nigeria UruguayEthiopia Norway VenezuelaGabon Oman YemenGhana Pakistan ZambiaGuatemala Papua New Guinea ZimbabweHonduras Paraguay

22

Table 3: Country’s specialization in commodity exports

Commodity weights in country’s total exportsCommodity 1 2 3Oil Nigeria Yemen Iran

95.12% 82.76% 79.72%Cotton Benin Mali Pakistan

61.00% 33.48% 20.52%Tobacco Malawi Zimbabwe Tanzania

60.50% 19.53% 6.35%Copper Zambia Chile Peru

59.99% 30.79% 13.93%Gold Mali Burundi Ghana

54.05% 35.45% 28.56%Coffee Burundi Ethiopia Uganda

50.98% 46.43% 36.87%Uranium Niger Benin Brazil

41.73% 29.90% 0.01%Cocoa Cote d’Ivoire Ghana Cameroon

34.10% 33.16% 9.75%Alu Mozambique Bahrain United Arab Emirates

33.44% 12.89% 12.77%Soya Paraguay Argentina Brazil

32.72% 4.45% 4.22%Fish Mauritania Mozambique Madagascar

30.96% 19.87% 14.34%Bananas Dominica Ecuador Costa Rica

29.20% 17.83% 12.83%Tea Kenya Sri Lanka Malawi

21.20% 15.12% 7.97%Crustaceans Mozambique Madagascar Nicaragua

18.96% 13.35% 10.67%Iron South Africa Zimbabwe Dominican Rep.

11.24% 10.45% 10.17%Sugar Guatemala Malawi Nicaragua

9.71% 9.15% 6.28%Rice Myanmar Uruguay Pakistan

8.99% 7.72% 7.08%Coal Colombia South Africa Indonesia

8.85% 6.24% 3.01%Beef Nicaragua Uruguay Paraguay

6.15% 4.78% 4.70%Tin Bolivia Peru Indonesia

6.03% 0.68% 0.64%Rubber Sri Lanka Thailand Indonesia

5.81% 5.10% 3.81%Notes.Weights are defined as the ratio between the commodity exports of thecountry and all the commodity exports of the country. Annual averages overthe period 1988-2008

23

Table 4: Final set of country-commodity pairs

Commodity Country Weight1 Oil Nigeria 95.1%2 Tobacco Malawi 60.5%3 Copper Zambia 60.0%4 Gold Mali 54.1%5 Coffee Burundi 51.0%6 Uranium Niger 41.7%7 Cocoa Cote d’Ivoire 34.1%8 Soya Paraguay 32.7%9 Fish Mauritania 31.0%10 Bananas Dominica 29.2%11 Tea Kenya 21.2%12 Crustaceans Mozambique 19.1%13 Alu Bahrain 12.9%14 Iron South Africa 11.2%Notes. Weights are defined as the ratio between the com-modity exports of the country and all the commodity ex-ports of the country. Annual averages over the period1988-2008. In the table, we have: 5 pairs for which themain commodity has a share larger than 50%, 6 pairslarger than 40%, 9 pairs larger than 30%, 11 pairs largerthan 10% and 14 pairs larger than 10%.

24

Table 5: Description of the commodity price seriesCommodity Source DescriptionAlu IFS Aluminum, LME standard grade, min-

imum purity, cif UK US$ per MetricTon

Bananas IFS Central American and Ecuador, firstclass quality tropical pack, Chiquita,Dole and Del Monte, U.S. importer’sprice FOB U.S. ports (Sopisco News,Guayaquil). $/Mt

Cocoa beans IFS International Cocoa Organization cashprice. Average of the three nearestactive futures trading months in theNew York Cocoa Exchange at noon andthe London Terminal market at clos-ing time, CIF U.S. and European ports(The Financial Times, London). $/Mt

Coffee (other milds) IFS International Coffee Organization,Other Mild Arabicas New York cashprice. Average of El Salvador centralstandard, Guatemala prime washedand Mexico prime washed, promptshipment, ex-dock New York. Cts/lb

Copper IFS London Metal Exchange, grade A cath-odes, spot price, CIF European ports(Wall Street Journal, New York andMetals Week, New York). Prior to July1986, higher grade, wirebars, or cath-odes. $/Mt

Crustaceans IFS Shrimp, U.S., frozen 26/30 count,wholesale NY US$ per pound

Fish IFS Fresh Norwegian Salmon, farm bred,export price (NorStat). US$/kg

Gold DS Gold Bullion LBM US$/Troy OunceIron IFS Iron Ore Carajas US cents per Dry

Metric Ton UnitOil IFS U.S., West Texas Intermediate 40o

API, spot price, FOB Midland Texas(New York Mercantile Exchange, NewYork). (In 1983-1984 Platt’s OilgramPrice Report, New York). $/bbl

Soya IFS Soybean U.S., cif Rotterdam US$ perMetric Ton

Tea IFS Mombasa auction price for best PF1,Kenyan Tea. Replaces London auctionprice beginning July 1998. Cts/Kg

Tobacco DS Tobacco, US (all markets), mid monthcurn

Uranium IFS Metal Bulletin Nuexco Exchange Ura-nium (U3O8 restricted) price. $/lb

Notes.“IFS” refers to International Financial Statistics from the IMF. “DS”refers to Datastream. “LME” refers to London Metal Exchange. “Cif” refersto cost, insurance and freight. “FOB” refers to free on board. “bbl” refers tobarrel (42 US Gallons). “API” refers to American Petroleum Institute.

25

Table 6: Basic descriptive statistics

Series min mean max std.dev ADF(p) pNigeria 3.4853 4.4486 6.0982 0.68673 -1.656 0Oil 4.2297 5.1401 6.3455 0.4649 -2.288 1Malawi 4.4747 5.1098 5.5274 0.29195 -2.002 1Tobacco 4.5073 4.8291 5.1419 0.13416 -2.376 1Zambia 3.241 4.6771 5.3882 0.31473 -2.573 1Copper 3.9584 4.4385 5.5381 0.35539 -1.863 8Mali 4.5016 5.0082 5.7875 0.41613 -1.748 7Gold 4.3503 4.7661 5.4543 0.24692 -1.875 1Burundi 4.1084 4.6251 5.2196 0.28362 -1.088 0Coffee 3.3921 4.272 5.2334 0.3995 -2.059 5Niger 4.2787 4.9547 5.7179 0.36193 -1.758 2Uranium 4.2447 5.1067 7.0311 0.60505 -1.189 10Cote d’Ivoire 4.3603 4.8014 5.0642 0.14849 -2.656 0Cocoa 4.204 4.8064 5.6788 0.35895 -1.862 2Paraguay 4.270 4.6535 5.2845 0.25216 -2.312 1Soya 4.5049 4.8607 5.2845 0.25216 -3.847** 1Mauritania 4.0602 4.6674 5.282 0.36798 -0.9941 0Fish 4.0098 4.7319 5.4332 0.36975 -1.634 6Dominica 4.3684 4.5486 4.7373 0.072899 -2.634 0Banana 4.3005 4.99 5.5221 0.22919 -3.050* 9Kenya 4.0221 4.5563 5.1507 0.19515 -0.3104 4Tea 4.4078 4.9566 5.9459 0.24846 -3.194* 2Notes. Series are in logarithm, normanlized and, as regards commodityprices, deflated by manufacture unit value index (MUV). Critical valuesfor the Augmented Dickey-Fuller tests (with constant) are -2.87 for 5%and -3.45 for 1%.

Table 7: Pesaran test of cross-sectional dependence for different panels of countries

# countries Pesaran Test p-value Dependence? AACR5 21.66 0.00 Yes 0.40610 36.929 0.00 Yes 0.42015 36.929 0.00 Yes 0.445

Notes. AACR stands for Average Absolute value of the off-diagonalelements of the Correlation matrix of Residuals. The Pesaran test andthe AACR are based on fixed effects models. Results from randommodels are equivalent (non reported here). The null of the Pesarantest is the absence of cross sectional dependence. The samples of 5,10 and 15 units are based on selection of commodity country pairsfor which the weight of the dominant commodity in the country totalexport is the largest.

26

Table 8: Subsampling-based panel unit root tests for commodity prices

Test test value block rule crit val (5%) block size crit val (10%) block sizeLLC -7.718 SC -8.940 68 -8.616 66

MV -8.665 32 -8.313 17IPS -2.638 SC -2.518* 48 -2.351* 32

MV -2.470 24 -2.351 32INVN -0.0360 SC -2.078 117 -2.109 143

MV -2.578 30 -2.349 29Notes. Commodity prices are in logarithms. An intercept and a trend have beenconsidered in all the experiments. “SC” and “MV” hold for Stochastic Calibration andMinimum Volatility, respectively, two alternative block selection rules. The minimumand maximum block sizes for the MV rule are 0.4 and 0.6, respectively. An asteriskindicates the rejection of the null of panel unit root. The 5% and 10% value columnsgive the non-centered critical values.

Table 9: Subsampling-based panel unit root tests for the real exchange rates

Test test value block rule crit val (5%) block size crit val (10%) block sizeLLC -6.479 SC -9.241 68 -7.785 66

MV -11.428 16 -10.157 17IPS -2.269 SC -2.726 48 -2.556 32

MV -2.848 20 -2.676 20INVN -0.5777 SC -3.055 117 -3.190 143

MV - 4.423 15 -4.062 15Notes. Real exchange rates are in logarithms. An intercept and a trend have beenconsidered in all the experiments. “SC” and “MV” hold for Stochastic Calibration andMinimum Volatility, respectively, two alternative block selection rules. The minimumand maximum block sizes for the MV rule are 0.4 and 0.6, respectively. An asteriskindicates the rejection of the null of panel unit root. The 5% and 10% value columnsgive the non-centered critical values.

Table 10: Fachin (2006) Panel Cointegration Tests

Test value p-valuesbasic bootstrap FDB1 FDB2

With common time dummiesMean CADF -3.25 2.00 3.00 4.00Median CADF -3.28 0.00 1.00 1.00

Without common time dummiesMean CADF -3.21 0.00 0.00 0.00Median CADF -3.22 1.00 0.00 1.00Notes. Real exchange rates and commodity prices are in logarithms.An intercept and a trend have been considered in all the tests. Blocksize selection for the cointegration test is based on 0.1T. “FDB1” and“FDB2” hold for Fast Double Boostraps of types 1 and 2 (see Fachin(2006)). P-values are in percent.

27

Table 11: Cointegration esimates by DOLS, FMOLS and BKN methods

Test β Standard errorsPanel DOLS 0.326* 0.0022Panel FMOLS 0.317* 0.0021BKN 0.149* 0.0092Notes. An intercept and a trend have been considered in all the tests.An asterisk indicates that the coefficients are significant at a level of5 percent. Panel DOLS are based on 4 leads and lags. Panel FMOLSbased on Fejer kernel with window length 3.21 ∗ T 1/3. BKN is theCupFM estimator (see Bai et al. (2009)). One factor was sufficientfor the BKN estimate, convergence occured after 6 iterations, withquadratic spectral kernel. The estimates are made for a sample ofcountries where the dominant commodity has a share of at least 20%in total exports.

28

Table 12: Fachin(2006)’s cointegration tests for different samples

Test Stat p-Btst FDB1 FDB2 Coint.?14 countries (> 10%)

With common time dummies Mean CADF -2.84 18 22 21 NoMedian CADF -2.7 34 42 43 No

Without common time dummies Mean CADF -3.04 2 2 3 yesMedian CADF -3.09 1 1 1 yes

12 countries (> 15%)With common time dummies Mean CADF -2.95 7 10 10 yes

Median CADF -2.65 37 37 41 NoWithout common time dummies Mean CADF -3.13 1 1 2 yes

Median CADF -3.22 0 0 0 yes11 countries (> 20%)

With common time dummies Mean CADF -3.29 0 0 -1 yesMedian CADF -3.23 2 3 4 yes



Median CADF -3.24 3 6 5 yesWithout common time dummies Mean CADF -3.42 0 0 0 yes

Median CADF -3.54 0 10 0 yes6 countries (> 40%)

With common time dummies Mean CADF -3.32 4 5 6 yesMedian CADF -3.16 4 5 5 yes



Median CADF -3.17 8 9 11 yesWithout common time dummies Mean CADF -3.74 0 0 0 yes

Median CADF -3.77 0 0 0 yesNotes. The sample of respectively 14, 12, 11, 9, 6 and 5 countries include countrieswhere the dominant commodity has a share of at least 10%, 15%, 20%, 30%, 40% and50% in total exports. The 3 p-values are referred to as p-Btst, FDB1 and FDB2. SomeFDB2 are negative, which is not due to computation errors but to the corrections inthe FDB2 formula. See Davidson and MacKinnon (2000) p.7 for comments relatedto this potential negativity.

29

Table 13: Cointegration tests and estimates for different samples

Weight of the main cdty # countries Cointegration tests Coefficients SEMore than 50% (5 countries) C 0.284 0.0131More than 40% (6 countries) C 0.242 0.0101More than 30% (9 countries) C 0.239 0.0088More than 20% (11 countries) C/NC 0.149 0.0092More than 15% (12 countries) NC (0.113) (0.0117)More than 10% (14 countries) NC (0.165) (0.0106)Notes. Countries in each panel are selected according to the weight of their main com-modity in their total commodity exports. Cointegration test refers to mean CADFand median CADF with common time dummies of Fachin (2006). We report coin-tegration (C) if at least 5 of the 6 p-values are not larger than 10%. We reportnon-cointegration (NC) if at least 5 of the 6 p-values are larger than 10%. Other-wise, we report C/NC. The coefficients are cointegration estimates obtained with themethodology of BKN. SE holds for standard errors.

Table 14: Cointegration esimates by the DOLS, FMOLS and BKN methods for differentsamples

# countries min weight PDOLS FMOLS BKN5 50% 0.532* (0.0032) 0.524* (0.0032) 0.284* (0.013)6 40% 0.391* (0.0030) 0.384* (0.0029) 0.242* (0.0101)9 30% 0.345* (0.0024) 0.336* (0.0024) 0.239* (0.0088)11 20% 0.326* (0.0022) 0.317* (0.0021) 0.149* (0.0092)

Notes. The different samples of respectively 5, 6, 9 and 11 countries are based oncommodity country pairs where the weight of the main commodity is at least 50%,40%, 30% and 20% of total exports. An intercept and a trend have been consideredin all the tests. An asterisk indicates that the coefficients are significant at a level of5 percent. Standard errors in parentheses

Table 15: List of outliers

Country Outliers variation MonthNigeria -65.3% 1986-10Zambia +133.7% 1987-5

-59.2% 1985-10+55.1% 1987-2+50.4% 1981-8

Mali +51.9% 1995-10Notes. List of monthly variations of the real exchangerates larger than 50%.

30

Table 16: Break robustness check - panel unit root tests for the real exchange rates

Removal of monthly variations larger than 50%Test test value block rule crit val (5%) block size crit val (10%) block sizeLLC -6.685 SC -8.579 68 -7.838 66

MV -8.960 28 -8.112 26IPS -2.089 SC -2.552 47 -2.354 31

MV -2.620 20 -2.375 26INVN -0.249 SC -2.741 115 -2.552 142

MV - 2.378 32 -2.122 32Removal of monthly variations larger than 50% and of January1994 variations in Mali, Cote d’Ivoire and NigerTest test value block rule crit val (5%) block size crit val (10%) block sizeLLC -6.653 SC -8.149 68 -7.516 66

MV -8.718 29 -8.039 26IPS -2.113 SC -2.531 47 -2.344 31

MV -2.626 19 -2.337 28INVN 1.273 SC -1.450 115 -1.338 142

MV - 2.382 31 -2.414 26Notes. Subsampling-based panel unit root tests. Real exchange rates are in loga-rithms. An intercept and a trend have been considered in all the experiments. “SC”and “MV” hold for Stochastic Calibration and Minimum Volatility, respectively, twoalternative block selection rules. The minimum and maximum block size for the MVrule are 0.4 and 0.6, respectively. An asterisk indicates the rejection of the null ofpanel unit root. The 5% and 10% value columns give the non-centered critical values.

Table 17: Break robustness check - Fachin (2006)’s panel cointegration tests

Removal of monthly variations larger than 50%Test value p-values

basic bootstrap FDB1 FDB2With common time dummies

Mean CADF -3.04 1.00 0.00 -3.00Median CADF -2.83 15.00 12.00 11.00

Without common time dummiesMean CADF -3.07 3.00 4.00 5.00Median CADF -2.91 8.00 5.00 6.00Removal of monthly variations larger than 50% andof January 1994 variations in Mali, Cote d’Ivoire andNiger

Test value p-valuesbasic bootstrap FDB1 FDB2

With common time dummiesMean CADF -3.08 2.00 0.00 2.00Median CADF -2.99 5.00 4.00 3.00

Without common time dummiesMean CADF -3.07 1.00 0.00 0.00Median CADF -2.91 7.00 5.00 3.00Notes. Real exchange rates and commodity prices are in logarithms.An intercept and a trend have been considered in all the tests. Blocksize selection for the cointegration test is based on 0.1T. “FDB1” and“FDB2” hold for fast double boostraps (see Fachin (2006)). P-valuesare in percent. See Davidson and McKinnon (2000) p.7 for commentsrelated to the potential negativity of FDB2.

31

Figure 1: Selected commodity prices series, normalized (1995=100) and deflated (by theMUV) - by row, from left to right: oil, tobacco, copper, gold, coffee, uranium, cocoa, soybean,fish, banana and tea.

Figure 2: Normalized (1995=100) Real Effective Exchange Rates - by row, from left to right:Nigeria, Malawi, Zambia, Mali, Burundi, Niger, Cote d’Ivoire, Paraguay, Mauritania, Do-minica and Kenya

32

Figure 3: MUV: Unit Value Index (in US dollars) of manufactures (commodity deflator) -normalized version (1995=100)

33

(a) Nigeria and oil (b) Malawi and tobacco

(c) Zambia and copper (d) Mali and gold

(e) Burundi and coffee (f) Niger and uranium

Figure 4: Real exchange rates (dashed lines) and commodity prices (solid lines). Both seriesare normalized (1995=100) and commodity price are deflated by the manufactures unit valueindex (MUV)

34

(a) Cote d’Ivoire and cocoa (b) Paraguay and soya

(c) Mauritania and fish (d) Dominica and banana

(e) Kenya and tea

Figure 5: Real exchange rates (dashed lines) and commodity prices (solid lines). Both seriesare normalized (1995=100) and commodity price are deflated by the manufactures unit valueindex (MUV)

35

Figure 6: BKN cointegration coefficient for different samples (depending on the weight of themain commodity)

36

ISSN 1379-244X D/2011/3082/007

real exchanges rates in commodity producing countries: a ...signi cant positive long-run impact on...

Documents